Ender Demirkaya
Ender Demirkaya
**What changed?** With current logic, we retry tasks if the target domain is not active (failed over). This causes retry storms and it's not fixable unless the domain fails over...
**What changed?** During polls corrupt workflow is detected. This change deletes corrupt runs to avoid polls to corrupt workflows **Why?** Having a lot of such workflows can cause availability issues....
**What changed?** Adding logs when persistence queries take long to understand what queries are causing long persistence ops **Why?** **How did you test it?** **Potential risks** **Release notes** **Documentation Changes**
Our organization (github.com/cadence-workflow) uses mixed MIT and Apache 2.0 licenses. To standardize our organization, we are changing the license of all of our repos to Apache 2.0.
Cadence has been using consistent hashing based host-shard management. However this logic comes up with significant drawbacks which come to surface at high scale. Issues include: - Shard Load Skew...
Up to 1000x more scalable timer callback service for scalable lightweight scenarios
Instead of activity heartbeats, we will track liveness of workers and retry activities when a corresponding worker is detected to be down. Users won't need to add any heartbeat for...
Provide an improved data model to significantly lower database CPU usage
Support following roles: - Cadence Admin (can perform any action for any domain) - Domain Admin (can perform any action for an authorized domains) - Domain Viewer (can perform limited...